Optimizing I/O via kernel changes

Hi List,

I've been asked the following question by our VP of Development.

Is there some kernel parameter (or parameters) that we can tweak to improve our I/O on a Linux Server??

Since I'm new to Linux, I figured someone one one of these lists might have a concise answer to that
question. Can we use the kernel tomincrease I/O rates, or are we bound by hardware??

Thanks,
Mark Gosselin
NetScout Systems
-
To unsubscribe from this list: send the line "unsubscribe linux-newbie" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.linux-learn.org/faqs

Re: Optimizing I/O via kernel changes

At 11:30 AM 5/11/2004 -0400, Gosselin, Mark wrote:
>Hi List,
>
>I've been asked the following question by our VP of Development.
>
>Is there some kernel parameter (or parameters) that we can tweak to
>improve our I/O on a Linux Server??
>
>Since I'm new to Linux, I figured someone one one of these lists might
>have a concise answer to that
>question. Can we use the kernel tomincrease I/O rates, or are we bound by
>hardware??

This question is too vague to elicit a meaningful answer. Even assuming you
mean Mbps on an Ethernet by "I/O rates" (there are other possible meanings,
of course), knowing whether your particular setup is hardware or software
bound requires knowledge of your hardware and software ...

what NIC? (if not 100 Mbps, mention speed)
what CPU? (type and speed)
how much RAM?
is the system *ever* using swap?
any relevant IRQ sharing?

what Linux distro and version?
what Linux kernel ("uname -a")?
custom or stock kernel?
what NIC module and parameters?
what (if any) iptables rules are you running?
does "ifconfig" show any appreciable error rate?

what relevant services are you running?
what "I/O rates" are you currently seeing?
if the relevant connectivity is not just Ethernet (for example, if
it is a DSL connection to the Internet), what type and speed of connection
is it?
how "busy" is the system during normal operation?
what % CPU use does "top" report?
what blocking ("load averages") does "uptime" report?

Depending on the details of the response, there could be ways to speed up
throughput, such as reducing the complexity of firewall rulesets, tweaking
NIC module settings, rearranging IRQ assignments, conceivably doing some
tailoring to specific service needs, possibly tailoring packet size (MTU)
settings to a link layer other than Ethernet.

As a general matter, you may want to look at the settings described in
your kernel Documentation under ./Documentation/networking/ip-sysctl.txt
and at the doc file in the same directory for your NIC. But there is no
magic "faster" switch in the kernel (or at least none I know of) that is
turned off by default (why would there be?).

In practice, 100 Mbps Ethernet NICs under Linux seem to deliver somewhere
between 50 Mbps and 80 Mbps on a sustained basis, depending on things I'm
not sure of (kernel version? specific NIC hardware?).

And, of course, if I have misinterpreted your use of "I/O rates", this
entire response is meaningless for your purposes.




-
To unsubscribe from this list: send the line "unsubscribe linux-newbie" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.linux-learn.org/faqs

RE: Optimizing I/O via kernel changes

Both Chris and Mark have replied to me privately on this, but replies to
the list are more useful.

I aldready discarded Mark's reply, so I cannot forward it ... but he said
he asked for more details and was told that the speedup issue involves SCSI
I/O of some sort ... I've forgotten the specifics ... a topic way outside
my expertise.

So while Chris was closer to the target than I, Mark really needs to post a
more specific follopup to the list in order to get help.

At 09:30 AM 5/12/2004 -0500, Little, Chris wrote:
>see, i took this to mean disk i/o. the next guy might take it as video i/o.
>vague questions lead to vague answers, eh?
>
> > -----Original Message-----
> > From: Ray Olszewski [mailto:ray@comarre.com]
> > Sent: Tuesday, May 11, 2004 11:52 AM
> > To: linux-newbie@vger.kernel.org
> > Subject: Re: Optimizing I/O via kernel changes
> >
> >
> > At 11:30 AM 5/11/2004 -0400, Gosselin, Mark wrote:
> > >Hi List,
> > >
> > >I've been asked the following question by our VP of Development.
> > >
> > >Is there some kernel parameter (or parameters) that we can tweak to
> > >improve our I/O on a Linux Server??
> > >
> > >Since I'm new to Linux, I figured someone one one of these
> > lists might
> > >have a concise answer to that
> > >question. Can we use the kernel tomincrease I/O rates, or
> > are we bound by
> > >hardware??
> >
> > This question is too vague to elicit a meaningful answer.
> > Even assuming you
> > mean Mbps on an Ethernet by "I/O rates" (there are other
> > possible meanings,
> > of course), knowing whether your particular setup is hardware
> > or software
> > bound requires knowledge of your hardware and software ...
> >
> > what NIC? (if not 100 Mbps, mention speed)
> > what CPU? (type and speed)
> > how much RAM?
> > is the system *ever* using swap?
> > any relevant IRQ sharing?
> >
> > what Linux distro and version?
> > what Linux kernel ("uname -a")?
> > custom or stock kernel?
> > what NIC module and parameters?
> > what (if any) iptables rules are you running?
> > does "ifconfig" show any appreciable error rate?
> >
> > what relevant services are you running?
> > what "I/O rates" are you currently seeing?
> > if the relevant connectivity is not just Ethernet
> > (for example, if
> > it is a DSL connection to the Internet), what type and speed
> > of connection
> > is it?
> > how "busy" is the system during normal operation?
> > what % CPU use does "top" report?
> > what blocking ("load averages") does "uptime" report?
> >
> > Depending on the details of the response, there could be ways
> > to speed up
> > throughput, such as reducing the complexity of firewall
> > rulesets, tweaking
> > NIC module settings, rearranging IRQ assignments, conceivably
> > doing some
> > tailoring to specific service needs, possibly tailoring
> > packet size (MTU)
> > settings to a link layer other than Ethernet.
> >
> > As a general matter, you may want to look at the settings
> > described in
> > your kernel Documentation under
> > ./Documentation/networking/ip-sysctl.txt
> > and at the doc file in the same directory for your NIC. But
> > there is no
> > magic "faster" switch in the kernel (or at least none I know
> > of) that is
> > turned off by default (why would there be?).
> >
> > In practice, 100 Mbps Ethernet NICs under Linux seem to
> > deliver somewhere
> > between 50 Mbps and 80 Mbps on a sustained basis, depending
> > on things I'm
> > not sure of (kernel version? specific NIC hardware?).
> >
> > And, of course, if I have misinterpreted your use of "I/O
> > rates", this
> > entire response is meaningless for your purposes.





-
To unsubscribe from this list: send the line "unsubscribe linux-newbie" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.linux-learn.org/faqs

RE: Optimizing I/O via kernel changes

OK, here's the final scoop on this one... I finally found someone willing to tell
me EXACTLY what they want to do. We run an application called snort (snort.org), and
we run upwards of 100 instances of the snort process to collect data from a network
segment. We want to optimize the I/O to the SCSI device, so that all of these processes
can write to the same device (into different files).

We're running Red Hat Enterprise ES 3.0, on a Dell 2650, kernel is 2.4.21-9.ELsmp,
it's a stock kernel, system has 8GB of RAM, no real swap activity...

Anyone have any ideas???

Thanks,
Mark Gosselin
NetScout Systems

-----Original Message-----
From: Ray Olszewski [mailto:ray@comarre.com]
Sent: Wednesday, May 12, 2004 1:41 PM
To: linux-newbie@vger.kernel.org
Subject: RE: Optimizing I/O via kernel changes


Both Chris and Mark have replied to me privately on this, but replies to
the list are more useful.

I aldready discarded Mark's reply, so I cannot forward it ... but he said
he asked for more details and was told that the speedup issue involves SCSI
I/O of some sort ... I've forgotten the specifics ... a topic way outside
my expertise.

So while Chris was closer to the target than I, Mark really needs to post a
more specific follopup to the list in order to get help.

At 09:30 AM 5/12/2004 -0500, Little, Chris wrote:
>see, i took this to mean disk i/o. the next guy might take it as video i/o.
>vague questions lead to vague answers, eh?
>
> > -----Original Message-----
> > From: Ray Olszewski [mailto:ray@comarre.com]
> > Sent: Tuesday, May 11, 2004 11:52 AM
> > To: linux-newbie@vger.kernel.org
> > Subject: Re: Optimizing I/O via kernel changes
> >
> >
> > At 11:30 AM 5/11/2004 -0400, Gosselin, Mark wrote:
> > >Hi List,
> > >
> > >I've been asked the following question by our VP of Development.
> > >
> > >Is there some kernel parameter (or parameters) that we can tweak to
> > >improve our I/O on a Linux Server??
> > >
> > >Since I'm new to Linux, I figured someone one one of these
> > lists might
> > >have a concise answer to that
> > >question. Can we use the kernel tomincrease I/O rates, or
> > are we bound by
> > >hardware??
> >
> > This question is too vague to elicit a meaningful answer.
> > Even assuming you
> > mean Mbps on an Ethernet by "I/O rates" (there are other
> > possible meanings,
> > of course), knowing whether your particular setup is hardware
> > or software
> > bound requires knowledge of your hardware and software ...
> >
> > what NIC? (if not 100 Mbps, mention speed)
> > what CPU? (type and speed)
> > how much RAM?
> > is the system *ever* using swap?
> > any relevant IRQ sharing?
> >
> > what Linux distro and version?
> > what Linux kernel ("uname -a")?
> > custom or stock kernel?
> > what NIC module and parameters?
> > what (if any) iptables rules are you running?
> > does "ifconfig" show any appreciable error rate?
> >
> > what relevant services are you running?
> > what "I/O rates" are you currently seeing?
> > if the relevant connectivity is not just Ethernet
> > (for example, if
> > it is a DSL connection to the Internet), what type and speed
> > of connection
> > is it?
> > how "busy" is the system during normal operation?
> > what % CPU use does "top" report?
> > what blocking ("load averages") does "uptime" report?
> >
> > Depending on the details of the response, there could be ways
> > to speed up
> > throughput, such as reducing the complexity of firewall
> > rulesets, tweaking
> > NIC module settings, rearranging IRQ assignments, conceivably
> > doing some
> > tailoring to specific service needs, possibly tailoring
> > packet size (MTU)
> > settings to a link layer other than Ethernet.
> >
> > As a general matter, you may want to look at the settings
> > described in
> > your kernel Documentation under
> > ./Documentation/networking/ip-sysctl.txt
> > and at the doc file in the same directory for your NIC. But
> > there is no
> > magic "faster" switch in the kernel (or at least none I know
> > of) that is
> > turned off by default (why would there be?).
> >
> > In practice, 100 Mbps Ethernet NICs under Linux seem to
> > deliver somewhere
> > between 50 Mbps and 80 Mbps on a sustained basis, depending
> > on things I'm
> > not sure of (kernel version? specific NIC hardware?).
> >
> > And, of course, if I have misinterpreted your use of "I/O
> > rates", this
> > entire response is meaningless for your purposes.





-
To unsubscribe from this list: send the line "unsubscribe linux-newbie" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.linux-learn.org/faqs
-
To unsubscribe from this list: send the line "unsubscribe linux-newbie" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.linux-learn.org/faqs